difference between "Runtime API Reference" and "Driver API Reference" in the Ref

dsant · March 5, 2009, 4:51pm

what is the difference between Runtime API Reference and Driver API Reference in the Reference Manual ?

for example, we have CudaMalloc (p 29) in one, cuMemAlloc (p 177) in the other, Both are : “Allocates count bytes of linear memory on the device and returns in *devPtr a pointer to the allocated memory” ?

dsant · March 5, 2009, 7:33pm

I have the answer : External Image

[indent]CUDA is composed of two APIs:

A low-level API called the CUDA driver API,
A higher-level API called the CUDA runtime API that is implemented on top of
the CUDA driver API.[/indent]
Source : FAQ or [url=“http://documen.tician.de/pycuda/_sources/faq.txt”]http://documen.tician.de/pycuda/_sources/faq.txt[/url]

[indent]Pittsburgh asked: Is cuMemAlloc() the same as cudaMalloc() ?

MisterAnderson42 replied : As far as I know, they are the same.[/indent]
source : [url=“http://forums.nvidia.com/lofiversion/index.php?t71057.html”]http://forums.nvidia.com/lofiversion/index.php?t71057.html[/url]

tmurray · March 5, 2009, 7:41pm

They’re far from the same (cudaMalloc calls cuMemAlloc, sure, but it does a bunch of other stuff too), but that’s neither here nor there…

MisterAnderson42 · March 5, 2009, 8:33pm

Well, at at abstract level they essentially do the same thing, that is all I meant. Of course cudaMalloc has all sorts of extra stuff for the whole implicit context creation setup and whatnot.

jack · March 5, 2009, 9:35pm

Also, if you program with the CUDA runtime, I believe you also need to either distribute the runtime with your application, or install the CUDA toolkit/sdk on any machines that want to run the application (though I can’t vouch for this, as I only use my own dev box for CUDA stuff.)